How new, longer-life chips can help stretch your budget

Luckily for IT decision-makers, the time between hardware upgrade cycles for data centers is rather long. Servers and the processors they house are generally built to last with enough adaptability to power businesses through the best part of a decade. Nevertheless, nothing lasts forever: Machines fail, parts break, warranties run out. On top of that, the needs of businesses inevitably change, with demands for more power, more capacity, more technology.

The benefits of an upgrade are hard to ignore.  

The tech industry is littered with horror stories however, from power outages to misconfigured servers and data breaches – and there is no getting around the fact that when upgrading data center hardware there is risk involved. With such long periods between refresh cycles, there is a sizable amount of pressure to get the purchasing right as these are costly upgrades, after all. 

It’s important to consider, therefore, the choices IT teams need to make and the trade-offs they have to weigh up when it’s time to make a change in the data center, particularly when it comes to processors. 

Server refresh

Servers are the lifeblood of data centers, powering all their services and housing their data, but it's inside these devices that we find the computational prowess: The chip. The server vendor has a large say in the chip you end up using, though most have flexible options and a choice of at least two. That’s not to say you can’t breathe new life into an older server with a new chip – many are dual sockets, after all. 

The period between upgrades has been steadily growing, however, and is longer than ever before. Part of the reason for this is necessity: As Gartner VP analyst Alan Priestley puts it, most organizations have the mindset ‘If it ain't broke, don't fix it’. Within that will be considerations around cost, a fondness for the existing hardware, a fear of breaking things, or a combination of all three.  

“It used to be life cycles in the data center might be 5 years, but they've probably pushed out now closer to seven,” Priestley says. “You deploy a workload and you keep running it. You only invest in new infrastructure when you deploy new workloads, or you really need more performance. That’s if what you’ve got doesn't deliver in terms of the demand at the moment. And the decision they have to think about there, when they extend life cycles, is warranty.”

“There are some circumstances where refreshes will happen earlier because of products coming out, spent warranty, inability to get access to support, and spare parts.”

IT decision-makers have one big decision to start with, according to Priestley: Whether they land in the cloud or on-premises. The cloud can feel like the safe route, as it offers a roster of infrastructure and services like renting virtual machines, or even CPUs. 

“However, if they start looking at on-prem data centers, they own and operate the data center and those services – they’ve got infrastructure deployed in there, running specific workloads. And most of that stock goes in and stays in for an extended period of time.” 

Cores for concern

When it comes to what processor is used in a product, vendors have varying approaches to how much choice a customer may have. Many offer off-the-shelf options with a predefined CPU, GPU and other components, particularly for smaller customers. For larger enterprise customers, there may be more choice over the configuration. Whichever is the case, it’s worth considering what that chip is when making the server purchase as core counts matter. 

Thanks to virtualization and dual-socket servers, many workloads today don't need a full CPU to run. They can be split into lots of smaller virtual CPUs and run workloads in a virtualized environment. Most legacy servers in corporate enterprise data centers will be dual-socket servers, should an organization want that option, so there is the possibility of two CPUs per server or one CPU per server – whichever suits their needs best.

When choosing what to install, cores per processor is an important consideration.

“There are arguments for higher-performance cores – we could say that gives you better provisioning – other times, you just want blocks of cores,” Priestley explains. “The value proposition here, from AMD’s perspective with Milan and Genoa, is their higher core counts. Genoa, for example, goes up to 96 cores, Intel, today, goes up to [64].” This means an organization can put at least 96 virtual machines on a single Genoa CPU, or more with techniques such as multi-threading and overprovisioning.

As Priestly points out, this means an IT decision maker “can buy fewer servers to deliver the same workload and [the] TCI is better”.

Efficiency is a key area for IT decision makers when it comes to hardware procurement, which gives AMD’s Milan processors a big advantage. Two Epyc 7542 32cs offer a 36% performance increase over two Xeon 8358 3cs, according to AMD – and that also includes 82% better energy efficiency. Powering your servers with 4th Gen AMD Epyc processors can help organizations run the same amount of workloads with fewer servers and save on capital expenditure and operating expenses. 

While there is no way to avoid hardware upgrades, there are innovative, future-proof technologies out there to help power your business for as long as possible. It’s about making the right decisions across the stack, which includes processors, core counts, and energy efficiencies. The right server is a big part of that, but the right CPU is also important.